What is Red teaming?
Red teaming for AI is a structured, adversarial test in which people deliberately try to make a system fail, misbehave or be misused, in order to find vulnerabilities, harmful outputs and weak safeguards before real users or attackers do. It is common for generative AI and agents.
Findings are recorded, rated and fed into fixes and retesting. Red teaming complements, rather than replaces, standard testing and ongoing monitoring.